Generate images with Auteo
Uses the auteo CLI (npm i -g @auteo/cli, then auteo auth login). If a command exits with code 3, follow the auteo-account skill to sign in first. MCP users can do the same with the auteo_* tools listed at the end.
1. Pick a model
auteo model list --type text_to_image --json # or --type image_to_image for edits
auteo model get <model> --json # every parameter, options and constraints
Always read model get before generating: aspect ratios and resolution tiers differ per model, and resolutionsByAspectRatio says which tiers each ratio allows.
Rough guide (check the live list, it changes):
| Need | Good first choice |
|---|---|
| Best overall quality, text in images | openai/gpt-image-2-6-sunburst, google/nano-banana-pro |
| Fast, cheap drafts | google/nano-banana, alibaba/z-image |
| Editing a photo with reference images | google/nano-banana-2, bytedance/seedream-5-5 |
| Free for Plus / Ultra subscribers | models with subscriberUnlimited: true (e.g. alibaba/z-image) |
2. Choose parameters
--aspect-ratiois required. Use what the user asked for; otherwise1:1for avatars and icons,16:9for banners and thumbnails,9:16for stories,3:4for portraits. Some models also offerauto.--resolutionis required when the model has tiers (1K/2K/4K). Higher tiers cost more. Default to1Kfor drafts; only go higher when the user wants print or large display, and say it costs more.--n(1–4) makes several variations; each one is charged.--image <path|https-url>(repeatable) adds reference images for editing or style. Local files are uploaded; the number allowed is inmodel get(inputAssetIds.max).- Write the prompt in English unless the user wants text in another language rendered in the image.
3. Check the price, then generate
auteo generate cost google/nano-banana-2 --aspect-ratio 16:9 --resolution 2K --json
auteo generate create google/nano-banana-2 --prompt "a red fox in fresh snow, cinematic light" \
--aspect-ratio 16:9 --resolution 2K --out ./images --json
--out <dir> waits for the result and downloads the files; the JSON lists files and outputs[].url. Without --out/--wait the command returns a requestId immediately; fetch it later with auteo generate wait <requestId> --out ./images.
Tell the user the file paths and the credits spent (cost).
When it fails
- Exit 4, not enough credits: stop and follow the auteo-account skill (tell the user the price and balance, link to top up). Do not lower settings on your own.
- Exit 2, invalid parameters: re-read
auteo model get <model>and fix the ratio / tier combination. - Exit 5, prompt rejected by moderation: rewrite the prompt without the flagged content; resending the same prompt fails again.
- Exit 6, limit: too many generations in progress or too many submissions; wait and retry.
- Timeout or network error during
create: do not resubmit blindly. Runauteo generate list --type image --limit 5 --json; if the request is not there, retry with--client-request-id <the id printed before>so it cannot be charged twice. status: failedin the result: reporterrorMessage; ifsuggestedModelis set, offer it to the user.
MCP tools
auteo_list_models, auteo_get_model, auteo_estimate_cost, auteo_upload_image (returns an assetId for inputAssetIds), auteo_generate_image, auteo_get_generation (call again after pollAfterMs until done), auteo_list_generations.